Edge-Constrained UAV Small-Object Detection with P2 Enhancement and Quantum-Inspired Lightweight Structure Search
Combining P2 enhancement and quantum-inspired algorithms, the method improves UAV small-object detection AP by 31.10%.
Key Findings
Methodology
This study builds on YOLOX-Nano, integrating a high-resolution P2 branch to preserve shallow spatial details. A quantum-inspired evolutionary algorithm (QIEA) guides lightweight structure search, with the search space designed around lightweight modules and task-specific parameters. Evaluation involves accuracy, FLOPs, latency, memory, and recall, using proxy training for rapid assessment. Results show P2 significantly boosts APsmall by 31.10%, with QIEA-selected structures achieving the highest recall. Multiple heuristic comparisons validate QIEA’s efficiency, and full training confirms +P2’s superior detection performance on VisDrone, with AP50:95 increased by 17.5% and APsmall by 44.9%.
Key Results
- Adding the P2 branch improved APsmall by 31.10% on VisDrone, demonstrating effective enhancement for small-object detection.
- YOLOX-Nano+P2 outperformed NanoDet-Plus with similar size, increasing AP50:95 by 17.5% and APsmall by 44.9%.
- QIEA-optimized structures showed the highest recall, but full training favored the +P2 model for overall accuracy.
Significance
This work addresses the critical challenge of small-object detection on resource-constrained edge devices, proposing an innovative combination of high-resolution feature enhancement and quantum-inspired structure search. It advances the state-of-the-art by enabling lightweight models to achieve high detection accuracy, crucial for UAV applications like surveillance and navigation. The comprehensive evaluation across multiple metrics and datasets demonstrates its practical potential, bridging the gap between detection performance and deployment feasibility. The approach paves the way for more intelligent, autonomous UAV systems capable of real-time perception in complex environments.
Technical Contribution
The paper introduces a high-resolution P2 branch into YOLOX-Nano, effectively preserving shallow spatial features vital for small-object detection. It also designs a quantum-inspired evolutionary algorithm (QIEA) for lightweight structure search, utilizing probabilistic encoding and proxy evaluation to efficiently explore the design space. This hybrid approach combines neural architecture search principles with quantum-inspired heuristics, enabling automatic, resource-aware optimization. The method balances detection accuracy with hardware constraints, providing a novel, automated framework for edge deployment of UAV detectors.
Novelty
This is the first integration of quantum-inspired algorithms into UAV small-object detection structure search, combined with high-resolution feature enhancement. Unlike traditional NAS or manual tuning, this approach employs probabilistic quantum-like encoding for efficient exploration under strict resource constraints, offering a new paradigm for lightweight detector design. Its focus on edge deployment and resource-aware optimization distinguishes it from existing methods that prioritize accuracy alone.
Limitations
- Proxy training metrics may not fully correlate with full training performance, leading to potential discrepancies in structure selection.
- Performance in extremely cluttered or tiny (<16 pixels) targets remains limited, requiring further multi-modal or multi-scale strategies.
- The search process, although efficient, still faces scalability issues in larger, more complex search spaces, necessitating more advanced optimization techniques.
Future Work
Future research will explore multi-modal data integration (e.g., infrared, LiDAR) to enhance detection robustness. Multi-task structure search aims to optimize detection alongside other UAV perception tasks. Additionally, incorporating hardware-aware multi-objective optimization could further improve deployment efficiency, enabling real-time, high-precision UAV perception in diverse scenarios.
AI Executive Summary
Unmanned aerial vehicles (UAVs) are increasingly deployed for tasks requiring real-time small-object detection under strict resource constraints. Traditional models often struggle to balance detection accuracy with hardware limitations, especially for tiny targets that demand high spatial resolution. Existing lightweight detectors either sacrifice shallow spatial details or rely on manual module addition, which can be unstable and costly.
This paper introduces a novel framework that combines a high-resolution P2 feature enhancement branch with a quantum-inspired evolutionary algorithm (QIEA) to automatically search for lightweight network structures optimized for edge deployment. The P2 branch preserves shallow spatial information crucial for small-object detection, significantly boosting APsmall by 31.10% on VisDrone. Meanwhile, QIEA efficiently explores the design space, selecting structures that balance accuracy, FLOPs, latency, and memory.
Experimental results demonstrate that YOLOX-Nano+P2 surpasses comparable models, achieving 17.5% higher AP50:95 and 44.9% higher APsmall over NanoDet-Plus with similar size. The QIEA-selected structure attains the highest recall, though full training shows +P2’s dominance in accuracy. Multiple validation strategies confirm the robustness of the approach, highlighting its potential for real-world UAV applications.
This work advances UAV perception by providing an automated, resource-aware structure optimization method that effectively enhances small-object detection. Its implications extend to surveillance, traffic monitoring, and disaster response, where resource efficiency and detection precision are paramount. Future directions include multi-modal data fusion, multi-task optimization, and hardware-aware design, aiming to further improve UAV autonomous perception in complex environments.
Deep Analysis
Background
随着无人机在监控、巡检等领域的广泛应用，目标检测技术不断发展。从两阶段的Faster R-CNN到单阶段的YOLO系列，模型在精度和速度上不断优化。边缘计算的兴起促使轻量化模型成为研究重点，MobileNet、EfficientNet等引领模型压缩和高效设计。小目标检测尤为依赖高分辨率特征和多尺度融合，但在复杂背景和密集场景中仍存在性能瓶颈。近年来，结合注意力机制、特征增强和结构搜索的研究不断推进，但在边缘设备上的部署效率和检测精度仍需提升。
Core Problem
无人机平台资源有限，难以部署复杂模型。小目标检测中，浅层空间信息丢失导致召回率低，手工设计结构难以兼顾性能与成本，缺乏自动化优化工具。如何在有限硬件条件下实现高效、准确的小目标检测，成为亟待解决的关键问题。现有方法多依赖经验调优或简单特征增强，缺乏系统化的结构搜索与资源优化机制。
Innovation
本文提出在YOLOX-Nano基础上引入P2高分辨率分支，增强浅层空间信息，显著提升小目标检测性能。同时，设计量子启发式算法（QIEA）进行轻量化结构筛选，利用概率编码和代理训练实现快速评估。创新点在于结合量子思想与神经结构搜索，提供一种低成本、高效的自动结构优化方案。该方法在保持模型轻量的同时，大幅提升APsmall和召回率，突破边缘场景的性能瓶颈。
Methodology
- �� 以YOLOX-Nano为基础，加入P2高分辨率分支，增强浅层空间信息。• 设计QIEA，用Q-bit概率编码和代理训练快速筛选候选结构。• 构建搜索空间，围绕轻量化模块（CA、CSA、融合模块）和超参数（小目标损失权重、中心采样半径）展开。• 采用多轮代理训练验证候选结构，结合多指标（APsmall、AP50:95、召回率）进行评价。• 最终筛选出性能优异的结构，进行全训练验证，确保实际效果。
Experiments
在VisDrone数据集上，训练100轮，验证模型在AP50:95、APsmall、召回率等指标上的表现。对比多种启发式算法（随机、遗传、模拟退火）筛选结构，验证代理指标的相关性。引入外部AU-AIR数据集，测试模型泛化能力。通过参数、FLOPs、延迟和内存等指标评估模型资源消耗，确保方案适合边缘部署。还进行了消融实验，验证P2分支和QIEA的贡献。
Results
在VisDrone上，加入P2分支使APsmall提升31.10%，显著改善小目标检测。YOLOX-Nano+P2在参数和FLOPs相似的NanoDet-Plus基础上，AP50:95提升17.5%，APsmall提升44.9%。QIEA筛选的结构在召回率方面表现最佳，但全训练后，+P2模型在检测准确率上更优。多启发式算法验证了QIEA的效率和效果，结构筛选的代理指标与实际性能存在一定偏差，但整体趋势一致。
Applications
该方法适用于无人机巡检、交通监控、灾害评估等场景，能在资源有限的边缘设备上实现高效小目标检测。未来可结合多模态信息（如红外、激光雷达）扩展应用范围，支持多任务联合优化，提升复杂环境下的检测鲁棒性。
Limitations & Outlook
代理训练指标与全训练性能存在偏差，需优化代理评估机制。模型在极端复杂背景和超小目标（<16像素）检测中仍表现不足，需结合多模态信息或更深特征。算法在超大搜索空间中效率有限，未来需引入更高效的搜索策略或多目标优化。
Plain Language Accessible to non-experts
想象你在厨房里做饭，要准备一道复杂的菜肴。传统方法是按照菜谱一步步操作，花费时间还可能出错。现在，有一种智能厨师助手，可以根据厨房的空间和材料，自动帮你挑选最合适的食材和做法。它会考虑你手头的材料、时间限制和口味偏好，自动优化菜谱。这个助手就像本文中的算法，能在有限的空间和时间内，找到最优的解决方案，帮助你做出美味又快捷的菜肴。它还会不断学习，变得越来越聪明，帮你节省更多时间和精力。这就像无人机在复杂环境中，利用有限资源，快速准确地找到目标，提升效率和效果。
ELI14 Explained like you're 14
想象你在学校里玩一个找宝藏的游戏，你需要在有限的时间内找到最多的宝藏。你可以用一些聪明的小策略，比如优先找那些看起来更有可能藏有宝藏的地方。这个游戏就像无人机检测目标一样，面对复杂环境和有限的时间、计算能力，它需要用聪明的方法快速找到目标。论文里的算法就像你的策略助手，帮你筛选出最有可能藏有宝藏的地点，还能不断学习，变得更聪明。它用一种特别的“量子”思维，像在玩概率游戏一样，快速试错，找到最优的方案。最终，你可以在最短时间内找到最多的宝藏，就像无人机在边缘设备上快速检测出所有小目标一样。这种聪明的策略让任务变得更简单、更快，也更节省资源。
Abstract
Unmanned aerial vehicle (UAV) object detection requires compact detectors that retain small-object details under onboard computation and memory constraints. Repeated downsampling inlightweight networks weakens shallow spatial information, while manually adding attention orfusion modules may increase cost without stable gains. This study analyzes YOLOX-Nano underedge-deployment constraints by combining a P2 high-resolution detection branch with a quantum-inspired evolutionary algorithm (QIEA) for lightweight structure screening. The search space isdefined by lightweight priority and task specificity, and the evaluation jointly considers accuracy,floating-point operations (FLOPs), latency, memory consumption, and recall. On VisDrone, theP2 branch increases APamall by 31.10% over the YOLOX-Nano baseline. Compared with NanoDet-Plus with similar model size, YOLOX-Nano+-P2 improves APs0.ss by 17.5% and APamal by 44.9%.The QIEA-selected candidate obtains the highest Recallso, but +P2 remains the strongest AP-oriented variant after full training. Full 100-epoch verification of Random-best, GA-best, andSA/QUBO-best candidates further shows that proxy rankings do not necessarily transfer to finalAPse9s. These results support using P2 as the main small-object enhancement path and QIEA as alightweight tool for candidate screening and accuracy-cost analysis. The source code, configurationfiles, diagnostic scripts, and summarized results are available at https://github.com/Ming23233/UAV-QIEA-Edge-Detection